Since the inception of autonomous driving, LiDAR (Light Detection and Ranging) has been at the forefront of technological discussions. Despite ongoing debates about the viability of vision-only systems, LiDAR continues to be adopted by numerous manufacturers. The advent of embodied intelligence has once again catapulted LiDAR to the center stage as the primary perception hardware. But why does LiDAR elicit both disdain and desire?

LiDAR is an active sensor that measures the distance to surrounding objects by emitting laser beams and receiving their reflected signals. Imagine shining a flashlight in the dark and "listening" for echoes to determine location. LiDAR emits light pulses and records the time interval (Time-of-Flight, ToF) from emission to reflection. Given the known speed of light, the longer the time, the greater the distance. Similar to bats using ultrasonic waves, LiDAR employs light waves. Most automotive LiDARs use the ToF method, measuring distance by emitting ultra-short laser pulses and timing their return. Another advanced technology, Frequency Modulated Continuous Wave (FMCW), continuously varies the emitted light's frequency, enabling both distance and relative speed measurements (akin to police radar for vehicles). Both have pros and cons: ToF boasts fast response times, high accuracy, and mature applications, while FMCW offers robust anti-interference and speed measurement but is more complex and costly.

To achieve a 360° field of view, LiDAR scans the surroundings in various ways. Current scanning solutions include mechanical rotating, hybrid, and pure solid-state. Mechanical rotating LiDAR (e.g., early Velodyne products) rotates a motor-driven scanning unit for comprehensive coverage but is bulky, heavy, and has a limited lifespan. Hybrid (semi-solid-state) LiDAR moves select optical components, like micro-mirrors (MEMS mirrors) or small-range rotating mirrors, balancing speed, size, and cost. Pure solid-state LiDAR has no moving parts, utilizing optical phased arrays (OPA) or Flash scanning. OPA adjusts the phase difference between multiple laser beams on a single chip, while Flash emits a wide laser light area, capturing the entire scene instantaneously. Pure solid-state LiDAR is compact, vibration-resistant, but technically demanding and still in the breakthrough stage.

Development History of LiDAR

LiDAR's application began in the 1960s, post-laser invention, and quickly found use in aerospace. The US Apollo 15 lunar mission in the late 1970s carried a laser altimeter for ranging. However, commercial and scientific adoption didn't gain traction until the late 1980s, used in 3D scanning and terrain mapping by aircraft and satellites. In the 21st century, with autonomous driving and robotics rising, LiDAR entered a period of accelerated development. In 2005, Velodyne introduced the first 64-line mechanical rotating LiDAR for the DARPA Grand Challenge, weighing 13 kilograms and costing around $80,000 each. Despite its performance, high costs and bulkiness spurred demand for smaller solid-state LiDAR.

The 2010s were a "Warring States period" for LiDAR technology. At the 2016 CES, Quanergy unveiled the solid-state LiDAR prototype S3 (advertised at $200). Two years later, China's RoboSense launched the first-generation semi-solid-state LiDAR M1, used in models like Lucid Air and XPeng G9, winning the CES Innovation Award for two consecutive years. Numerous startups emerged, some thriving (like RoboSense), while others faltered due to technical or financial issues (like Quanergy, which bet on OPA but faced production limitations). During this decade, LiDAR technology flourished, with MEMS mirrors, micro-mirrors, OPA, Flash, and other technologies leading the way. However, mass-producible products focused mainly on mature routes or transitional solutions.

Entering the 2020s, autonomous driving commercialization surged LiDAR demand. Since 2018, about 120 mass-produced vehicle models worldwide have integrated LiDAR. Chinese automakers pioneered expanding LiDAR from high-end to ordinary passenger vehicles, driving a 68% year-on-year increase in the ADAS market size in 2024. Forecasts suggest global automotive-grade LiDAR installations will exceed 3 million units in 2025. Chinese manufacturers have risen swiftly, achieving mass production. In December 2024, Hesai delivered over 100,000 units in a single month for the first time; RoboSense shipped over 150,000 units in Q4 2023, totaling nearly 250,000 units for the year. As of 2024, four Chinese companies—Hesai, RoboSense, Huawei, and Innovusion—controlled about 88% of the global automotive-grade LiDAR market share.

Main Types of LiDAR

Classification by Ranging Principle: Time-of-Flight (ToF) vs. Frequency Modulated Continuous Wave (FMCW)

ToF emits ultra-short laser pulses and measures round-trip time for distance calculation, offering high accuracy and ease of implementation. FMCW linearly varies laser frequency, measures signal frequency offset, and directly calculates distance and speed, similar to police radar. Most automotive LiDARs use ToF, while some systems experiment with FMCW for stronger anti-interference and direct speed measurement.

Classification by Scanning Method: Mechanical Rotating vs. Semi-Solid-State vs. Fully Solid-State

Mechanical rotating LiDAR (e.g., early Velodyne products) rotates a motor-driven device for 360° scanning, providing wide angle coverage but is bulky and prone to wear. Semi-solid-state (hybrid) LiDAR moves select components (like micro-mirrors), enabling high-speed scanning with reduced size and cost. Fully solid-state LiDAR has no moving parts, using MEMS micro-mirrors, OPA, or Flash for instantaneous scene emission. OPA changes beam direction by adjusting multiple laser emitting elements' phase on a single chip; Flash LiDAR is like a "flashlight" capturing environmental information all at once. Strictly speaking, only OPA, Flash, and their improved versions (like two-dimensional addressable flash) belong to fully solid-state LiDAR.

Classification by Application Scenario: Short-Range/Mid-Range/Long-Range LiDAR

Short-range LiDAR typically detects tens of meters, covering close-range vehicle blind spots; mid-range LiDAR detects hundreds of meters, useful for urban environments; long-range LiDAR detects several hundred meters or more, common in highway scenarios. Designs balance detection distance and field of view through factors like laser power, receiver sensitivity, and lens size. Advances in lasers and detection chips enable these classifications within the same product line, e.g., a single LiDAR can meet mid- and long-range detection needs through multiple vertical laser sources.

Performance Parameters

Detection Distance

Related to laser power and target reflectivity, modern automotive-grade LiDAR can exceed 300 meters under ideal conditions. 955nm wavelength LiDAR is common in ADAS, offering low cost but limited distance; 1550nm wavelength LiDAR performs better in rain and fog, with a maximum detection distance over 300 meters but at a higher cost.

Field of View (FOV)

Rotating LiDAR achieves a 360° horizontal FOV, with vertical FOV typically ranging from 30° to 60°, depending on design. Fully solid-state LiDAR, lacking mechanical rotation, can be designed with discontinuous narrow horizontal angles, mostly covering a wide range directly ahead for forward perception.

Angular Resolution and Number of Lines

Horizontal angular resolution, affected by scanning frequency, can reach 0.01°; vertical resolution depends on the number of laser lines (16, 32, 64, 128, 192, or more). More lines increase vertical point cloud density, capturing finer 3D details.

Point Cloud Rate

Refers to points collected per unit time, with modern LiDAR output reaching tens of millions of points per second. This indicator, influenced by line count and rotation rate, directly relates to scan update rate and environmental reconstruction accuracy. The latest LiDAR can achieve up to 24 million points per second, significantly enhancing moving object and complex environment perception.

Ranging Accuracy

Typical automotive LiDAR achieves centimeter-level accuracy (errors of a few centimeters), superior to visual cameras and meeting safety warning needs. Accuracy is mainly affected by laser pulse width and clock resolution.

Other Parameters

Include laser safety class (mostly Class 1, completely safe for humans), power consumption, lifespan, etc. Mechanical LiDAR typically has a lifespan of thousands of hours, while solid-state LiDAR can reach tens of thousands or even hundreds of thousands of hours.

Application of LiDAR in Autonomous Driving and Embodied Intelligence

LiDAR is invaluable in autonomous driving and intelligent robots, generating high-precision 3D point clouds to identify obstacles, pedestrians, road signs, etc., providing data for path planning. In Advanced Driver Assistance Systems (ADAS), LiDAR is used for Automatic Emergency Braking (AEB), Forward Collision Warning (FCW), and Adaptive Cruise Control (ACC), enabling precise distance monitoring and gradual deceleration. Since 2018, about 120 mass-produced vehicle models worldwide have integrated LiDAR. Chinese manufacturers have expanded LiDAR application in the passenger vehicle market, driving a 68% year-on-year increase in ADAS sector shipments in 2024.

In fully autonomous driving (Robotaxi), LiDAR is a core sensor. Its unrestricted performance under various lighting conditions (night performance remains unchanged) and ability to provide high-resolution 3D scenes in real-time make it essential. Many autonomous driving companies (like Waymo and Baidu Apollo) equip test vehicles with multiple LiDAR units. It's widely acknowledged that achieving high-level autonomous driving in complex urban environments often requires multiple LiDAR units to cover the surroundings.

An autonomous driving test vehicle equipped with a rotating LiDAR (white cylindrical object on top) and cameras. The high-density point cloud generated by LiDAR helps the vehicle "see" obstacles ahead and around, achieving global environmental perception.

The rapid development of embodied intelligence technology has further fueled LiDAR demand. Many service robots, logistics handling robots, and cleaning robots (like robotic vacuum cleaners) use LiDAR for environmental scanning and localization (SLAM), ensuring safe and efficient navigation. RoboSense notes that its products have served over 2,800 robot customers, used in scenarios like drone aerial surveying, automated warehousing, and security inspection. With decreasing costs and increasing mass production, LiDAR is transitioning from a luxury vehicle option to the mass market. Future LiDAR will collaborate with cameras and millimeter-wave radars to construct multi-sensor fusion systems, enhancing perception capabilities through deep learning and expanding to smart cities, drone surveying and mapping, and automated agriculture.

Overview of Major Chinese LiDAR Manufacturers

Hesai Technology

Established in 2014 with its headquarters in Shanghai, Hesai stands as one of the pioneering domestic enterprises in the automotive LiDAR sector. In February 2023, Hesai made its debut on the Nasdaq in the United States, securing its status as the "first Chinese LiDAR stock." The company's diverse product portfolio encompasses short, medium, and long-range LiDAR solutions, which are widely integrated into ADAS and autonomous vehicles. According to prospectus data, Hesai shipped 2,900, 4,200, and 14,000 LiDAR units in 2019, 2020, and 2021, respectively. Following the launch of the semi-solid-state AT128 and the blind spot-filling F series in 2022, its annual sales volume soared to 80,400 units. Remarkably, in December 2024, Hesai achieved a milestone by delivering over 100,000 units in a single month, becoming the first automotive LiDAR company globally to exceed this threshold. Hesai boasts in-depth collaborations with over 20 domestic and international automakers, including BYD and Lixiang, and its products have been designated for numerous intelligent vehicle models. Notably, Hesai is actively expanding into overseas markets, collaborating with international automakers such as Mercedes-Benz and BMW.

RoboSense

Founded in Shenzhen in 2014, RoboSense specializes in LiDAR and robotic perception technology. According to the company's official website, RoboSense has emerged as the global market leader in LiDAR, providing sensors and solutions to over 2,800 customers in the robotics and related industries, as well as to more than 310 automakers. On January 5, 2024, RoboSense successfully listed on the main board of the Hong Kong Stock Exchange under the stock code 2498.HK. RoboSense's product offering spans short to long distances, featuring the mechanical rotating M series (M1/M2/MX, etc.) and the all-solid-state E series (E1/E2, etc.). The company witnessed a significant surge in business in the second half of 2023: monthly sales exceeded 30,000 units for the first time in October, with cumulative sales of over 150,000 units in the fourth quarter and nearly 250,000 units for the entire year. Additionally, RoboSense has developed its own automotive-grade LiDAR-specific SoC chip and large-area array detection chip for its all-solid-state product line, marking a significant advancement in domestic production capabilities.

Overall, Chinese manufacturers are swiftly ascending in the LiDAR industry, assuming a dominant position globally. Market analysis indicates that as of 2024, four Chinese manufacturers—RoboSense, Huawei, Hesai Technology, and Innovusion—collectively hold approximately 88% of the global automotive LiDAR market. With the expansion of mass production and the decreasing costs, the competitive edge of Chinese LiDAR enterprises is becoming increasingly pronounced. In the coming years, it is anticipated that smart cars will come equipped with LiDAR as a standard feature, heralding the industry's entry into the "era of mass production." Furthermore, relevant technologies will continue to undergo optimization and upgrades, paving the way for broader application prospects in autonomous driving and intelligent robots.

-- END --